中文摘要：

异构网络中TCP丢包区分机制对无线网络的稳定性起着重要的作用,包对探测包的单向传输时延（ROD）作为区分参数对不同丢包类型进行区分,算法的准确度依赖于ROD样本的隶属度函数及其参数估算。为了更加准确地区分丢包类型,通过对传统高斯混合模型的EM算法进行分析,提出了基于势函数的初始化方法,并且在网络拥塞和无线误码同时存在的情况下,将改进的EM算法（PEM）应用于不同丢包模式下隶属度函数的构建中。仿真验证了该算法具有较好的收敛特性和稳定性,并且对不同丢包模式隶属度函数的确定达到了很好的构建效果。

英文摘要：

The packet loss differentiating mechanism of TCP for heterogeneous networks plays an important role in the stability of wireless networks,relative one-way delay（ROD） of packet pair is used as the differentiating parameter to distinguish the loss type.The accuracy of this algorithm depends on ROD samples membership functions and parameters estimation.In order to differentiate the packet loss pattern more accurately,the initialization method based on potential functions is proposed by analyzing the traditional expectation maximization（EM） algorithm in Gaussian mixture model.Then the improved EM（PEM） algorithm is applied in the construction of different packet loss membership functions in the situation when the network congestion and wireless error coexist.The simulation results indicate that this algorithm has better convergence characteristics and stability,and has well building effect in the construction of different packet loss pattern membership functions.